Best dictation software for Linux in 2026: the paid apps don't ship here

Handy leads that group on community with 29,917 GitHub stars, SpeakoFlow offers the widest choice of speech models, and OpenWhispr sits between the two. Three more projects come up constantly in Linux dictation threads and belong in a different category: Buzz, whisper.cpp and nerd-dictation.

That absence is the whole story of this search result. Look for dictation software on Linux and you find forum threads, half-finished scripts, and pages about Mac apps. The reason is commercial rather than technical.

Why can't you just buy Wispr Flow or Superwhisper for Linux?

Because neither company builds a Linux version, and neither hides it. I checked all five of the popular paid or Mac-only apps against their own live pages on 19 August 2026. Not one of them offers a Linux download, a Linux beta, or a Linux waiting list. Searching their names alongside the word Linux is a dead end by design.

Here is what each vendor states on its own site. Wispr Flow says its dictation product runs on "Mac, Windows, iOS, and Android", and prices Pro at $15 per user per month billed monthly or $12 per user per month billed annually, with a free tier capped at 2,000 words per week on desktop. Superwhisper offers macOS 13.3 or later, Windows 10 or later on x64 and ARM64, and iOS 18 or later. MacWhisper is Mac only and sells a one-time Pro licence at €64. VoiceInk states plainly that it "works only on Apple Silicon Macs" and requires macOS 14.4 or later, priced from $29 lifetime for one Mac. Aqua Voice offers a Mac and an iOS download, with Pro at $8 per month.

Which Linux dictation tools actually work in 2026?

Six projects are worth your time, and they split into two groups. Handy, SpeakoFlow and OpenWhispr are desktop apps with a global shortcut that types into whatever window has focus. Buzz, whisper.cpp and nerd-dictation are something else: a transcription workbench, a library, and a scriptable one-file utility. Confusing the two groups is the most common mistake in Linux dictation advice.

Every figure below came from the project's own repository through the GitHub API on 19 August 2026.

• Handy. MIT, 29,917 stars, release v0.9.5 published 8 August 2026. Windows, macOS on both Intel and Apple Silicon, and x64 Linux. Ships .deb, AppImage and .rpm for x86_64 and aarch64, which is more Linux packaging than anyone else here. Default shortcut on Linux is Ctrl+Shift+D. Runs Whisper Small through Large, or Parakeet V3 on CPU, which Handy's own README measures at "~5x real-time speed on mid-range hardware (tested on i5)".
• SpeakoFlow. MIT, 126 stars, release v1.3.0 published 19 August 2026. Windows, macOS, Linux. AppImage and .deb for x86_64 and ARM64, plus an AUR package. Built on Handy's dictation core and adds a wider model list, on-device translation, and an assistant panel.
• OpenWhispr. MIT, 5,555 stars, release v1.8.3 published 13 August 2026. Genuinely cross-platform, with .deb, .rpm, AppImage and a tarball in the same release. Local Parakeet or Whisper, or cloud models with your own key.
• Buzz. MIT, 21,031 stars, release v1.4.4 published 14 March 2026. Available on Linux as a Flatpak, a Snap, or pip install buzz-captions. It does live microphone transcription, but into its own window. Its feature list has no global shortcut that types into another app, so it is a transcription tool rather than a dictation layer.
• whisper.cpp. MIT, 53,025 stars, release v1.9.2 published 4 August 2026. The most starred project in this article by a wide margin, and the engine several of the others are built on. It is a C and C++ port of Whisper with a command line front end, so it is what you reach for when you are building something, not when you want to dictate an email.
• nerd-dictation. GPL-3.0, 1,914 stars, no tagged release, last commit 10 October 2025. Describes itself as "Offline Speech to Text for Desktop Linux" and is a single Python script needing Python 3.6 or newer, the VOSK API, and an input simulation tool. Its own README notes that "Text from VOSK is all lower-case", and its roadmap still lists Windows and macOS as a maybe.

"Accessibility tooling belongs in everyone's hands, not behind a paywall." Handy's README, quoting handy.computer, on why the project is free.

How accurate is offline dictation on Linux?

Accurate enough that the model is no longer the weak link. The models these tools run are published, benchmarked by third parties, and identical whichever operating system loads them. NVIDIA Parakeet TDT 0.6B v3, the default in SpeakoFlow and an option in Handy and OpenWhispr, records 6.34% average word error rate and 1.93% on LibriSpeech test-clean.

Those figures come from the Open ASR Leaderboard and NVIDIA's own model card, not from any testing I ran. Because the model runs locally, its published accuracy is the accuracy you get, before any optional cleanup step.

Speed is where the choice of model shows up in daily use. The ASR Leaderboard paper (arXiv:2510.06961) records Parakeet TDT 0.6B v3 and Whisper Large v3 at the same 4.81% average WER on multilingual short-form audio, but at 1720 and 111 RTFx respectively. RTFx is seconds of audio processed per second of compute, so that is the same measured accuracy at roughly 15 times the throughput.

On Linux, SpeakoFlow accelerates transcription with OpenBLAS and Vulkan, falling back to CPU when no usable GPU is present. Handy's README recommends an Intel, AMD or NVIDIA GPU for its Whisper models on Ubuntu 22.04 or 24.04, and notes that Parakeet V3 runs CPU-only from Intel Skylake or equivalent AMD chips upward.

How do you install SpeakoFlow on Ubuntu, Arch, or Fedora?

Three routes, depending on your package manager. Arch users install from the AUR. Debian and Ubuntu users install the .deb, which registers a desktop entry and an icon. Everyone else, Fedora and openSUSE included, runs the AppImage. Downloads live on the releases page, and the current release is v1.3.0.

1. Arch, Manjaro, EndeavourOS. Install speakoflow-bin from the AUR with yay -S speakoflow-bin or paru -S speakoflow-bin. Your AUR helper handles updates from then on.
2. Debian 13, Ubuntu 24.04 and newer, Mint 22 and newer, Pop!_OS, Tuxedo OS. Download SpeakoFlow_1.3.0_amd64.deb, or SpeakoFlow_1.3.0_arm64.deb on ARM, then run sudo apt install ./SpeakoFlow_1.3.0_amd64.deb. The .deb is built on Ubuntu 24.04 and needs that era of glibc, so on Ubuntu 22.04 or Debian 12 use the AppImage instead.
3. Fedora, openSUSE, anything else. Download SpeakoFlow_1.3.0_amd64.AppImage or SpeakoFlow_1.3.0_aarch64.AppImage, make it executable with chmod +x, and run it. There is no .rpm yet, because the packaging does not bundle the speech engine correctly and an .rpm that installs but cannot transcribe would be worse than none. Handy and OpenWhispr both ship real .rpm builds if that matters more to you than anything else in this article.
4. Give the AppImage a menu entry. A single-file AppImage cannot register an icon or desktop entry by itself. Gear Lever or AppImageLauncher add that if you want it.
5. Pick a model, then set your shortcut. A short setup wizard offers the transcription models on first launch. The Linux default for dictation is Ctrl+Space, and the assistant panel is Ctrl+Alt+Space. Both are rebindable.

Push to talk is the default behaviour: hold the shortcut, speak, and release to have the text typed out. If holding a key while talking is awkward, switch Recording behavior to Tap in Settings, where one press starts and the next press stops. That setting applies to every recording shortcut at once.

Does Linux dictation work on Wayland or only X11?

It works on both, but Wayland is the harder case and every project handles it differently. Wayland deliberately gives an application no standard way to inject keystrokes into another window or to raise itself above other windows. Those are the two things a dictation tool needs.

Handy's README is direct about the cost: it lists "Limited support for Wayland display server" under known issues, asks you to install wtype or dotool for text input, and warns that on Ubuntu 26.04, which defaults to Wayland, wtype does not work and you need ydotool configured through systemd. nerd-dictation defaults to xdotool, which its documentation marks as "Compatible with the X server only", and offers DOTOOL, YDOTOOL or WTYPE for Wayland instead.

SpeakoFlow's constraint is its recording overlay, which has to float above everything else. Per the repository, that needs either the wlr-layer-shell protocol, used by wlroots compositors such as Sway and Hyprland and by KDE Plasma, or classic X11 keep-above stacking. A native GNOME session on Wayland supports neither, because Mutter does not implement wlr-layer-shell. So SpeakoFlow detects GNOME on Wayland and runs itself under XWayland, where keep-above works. X11 sessions and KDE or wlroots Wayland sessions already work without that. Two environment variables override the behaviour if you want to test it: SPEAKOFLOW_ALLOW_WAYLAND=1 forces native Wayland, and SPEAKOFLOW_NO_GTK_LAYER_SHELL=1 disables layer shell.

How does SpeakoFlow compare with Handy on Linux?

They share a foundation, so the comparison is narrow. SpeakoFlow is built on the open-source Handy dictation core by CJ Pais, so the transcription path started as the same code. The differences are packaging, how many models you can pick from, what happens after the transcript exists, and how much community sits behind each project.

Frequently asked questions